Sex differences in the song circuit and song acoustic complexity in male and female house wrens

In this study, we compared neural song circuit morphology to singing behavior recorded in the field for 17 male and 18 female house wrens. The acoustic complexity of house wren songs was quantified using a recently published machine learning approach. This data set includes recordings of all house wren songs used in this analysis along with Raven selection tables defining the boundaries of each syllable. This includes 109 female songs. R code used to extract acoustic features and estimate element diversity and our proxy for song acoustic complexity are included. Summaries of acoustic variables for each song and each element are provided as well as files necessary to replicate the analysis. For each bird, we measured volume, cell number, cell density, and neuron soma size for three song circuits, Area X, HVC (used as a proper name), and the robust nucleus of the arcopallium (RA), and one control region, the nucleus rotundus (Rt). This data set includes these neural morphology measurements for each bird as well as R code used to (1) compare males and females for each neural measurement and (2) explore the relationship between acoustic complexity and neural morphology within each sex.     

Methods
Wild house wrens were recorded in the field singing spontaneously or in response to playback recordings of male or female house wren songs. Songs were clipped from much longer song recordings with 1 second before the start and 1 second after the end of the song. No further processing occurred. All songs used in this analysis can be found in the "songs.zip" file. The start and end of each element in the song were defined manually in Raven using both the spectrogram and waveform. These boundaries can be found in the .txt file associated with each sound file (.wav file) in the "songs.zip" folder.

Signal-to-noise ratios (SNR) were used to select songs of suitable quality for the rest of the analysis. Users can use the "snr.and.automatic.frequency.detection.r" script to replicate this calculation for all sounds in the "songs.zip" file.  When songs with a suitable SNR were selected, we used this same R script to automatically detect the frequency boundaries of each element. These were then viewed in Raven and corrected for any obvious deviations driven by interfering background noise. These final values are included in the .txt file for each sound.

We then used a machine-learning approach to quantify the acoustic complexity of each song. After transforming and removing any colinear variables, an unsupervised random forest was used to determine which variables best divide the data. This results in a dissimilarity matrix for each syllable which was then transformed into vectors using classical multidimensional scaling. These vectors are "acoustic space" occupied by house wren song elements. A 95% minimum convex polygon was then used to determine how much acoustic space elements within a single song occupy. Songs that occupy more space have a larger range of signal types. This final calculation is referred to as element diversity and is our measure of acoustic complexity. The "snr.and.automatic.frequency.detection.r" script provides the workflow to replicate this acoustic complexity calculation starting with the songs and .txt files in the "songs.zip" file. Users may also skip earlier steps of this analysis by using the files "acoustic.parameters.csv", "Transformed.non-colinear.acous.meas.csv" or "mds.acoustic.area.points.csv" as described in the "README.md" document.

17 male and 18 female house wrens were collected, brains were removed, frozen, sectioned, and stained, and neural morphology was measured under brightfield microscopy to quantify neural morphology in three song control regions, Area X, HVC, and RA, and one control region, Rt. Further detailed methods can be found in the associated manuscript.  All neural morphology measurements can be found in "all.bird.neural.data.csv". The "statistics.and.figs.r" script provides the workflow to replicate all statistics and figures in the manuscript.  Here we compare males and females for each morphology metric and investigate how song acoustic complexity relates to neural morphology for each sex separately.